2020-02-11 14:02:55

by Joerg Roedel

Subject: [PATCH 46/62] x86/sev-es: Handle INVD Events

From: Tom Lendacky <[email protected]>

Implement a handler for #VC exceptions caused by INVD instructions.

Signed-off-by: Tom Lendacky <[email protected]>
[ [email protected]: Adapt to #VC handling infrastructure ]
Co-developed-by: Joerg Roedel <[email protected]>
Signed-off-by: Joerg Roedel <[email protected]>
---
arch/x86/kernel/sev-es.c | 8 ++++++++
1 file changed, 8 insertions(+)

diff --git a/arch/x86/kernel/sev-es.c b/arch/x86/kernel/sev-es.c
index e96332516c2a..485f5a14c3b4 100644
--- a/arch/x86/kernel/sev-es.c
+++ b/arch/x86/kernel/sev-es.c
@@ -308,6 +308,11 @@ static enum es_result handle_rdpmc(struct ghcb *ghcb, struct es_em_ctxt *ctxt)
return ES_OK;
}

+static enum es_result handle_invd(struct ghcb *ghcb, struct es_em_ctxt *ctxt)
+{
+ return ghcb_hv_call(ghcb, ctxt, SVM_EXIT_INVD, 0, 0);
+}
+
static enum es_result handle_vc_exception(struct es_em_ctxt *ctxt,
struct ghcb *ghcb,
unsigned long exit_code,
@@ -328,6 +333,9 @@ static enum es_result handle_vc_exception(struct es_em_ctxt *ctxt,
case SVM_EXIT_RDPMC:
result = handle_rdpmc(ghcb, ctxt);
break;
+ case SVM_EXIT_INVD:
+ result = handle_invd(ghcb, ctxt);
+ break;
case SVM_EXIT_CPUID:
result = handle_cpuid(ghcb, ctxt);
break;
--
2.17.1

2020-02-12 00:12:45

by Andy Lutomirski

Subject: Re: [PATCH 46/62] x86/sev-es: Handle INVD Events

> On Feb 11, 2020, at 5:53 AM, Joerg Roedel <[email protected]> wrote:
>
> From: Tom Lendacky <[email protected]>
>
> Implement a handler for #VC exceptions caused by INVD instructions.

Uh, what? Surely the #VC code can have a catch-all OOPS path for things like this. Linux should never ever do INVD.

2020-02-12 15:36:45

by Joerg Roedel

Subject: Re: [PATCH 46/62] x86/sev-es: Handle INVD Events

On Tue, Feb 11, 2020 at 04:12:19PM -0800, Andy Lutomirski wrote:
>
>
> > On Feb 11, 2020, at 5:53 AM, Joerg Roedel <[email protected]> wrote:
> >
> > From: Tom Lendacky <[email protected]>
> >
> > Implement a handler for #VC exceptions caused by INVD instructions.
>
> Uh, what? Surely the #VC code can have a catch-all OOPS path for things like this. Linux should never ever do INVD.

Right, its hard to come up with a valid use-case for INVD in the Linux
kernel. I changed the patch to mark INVD as unsupported and print an
error message.

Regards,

Joerg